View previous topic :: View next topic |
Author |
Message |
Kenneth_Smith
Joined: 18 May 2012 Posts: 697 Location: Hamilton, Lanarkshire, Scotland.
|
|
Back to top |
|
|
John-Silver
Joined: 30 Jul 2013 Posts: 1520 Location: Aerospace Valley
|
Posted: Mon May 15, 2017 8:17 am Post subject: |
|
|
interesting - I wonder if it's just a PR exercise with secondary intent of sourcing some bright oprogrammers ?
$55000 is peanuts to them and they want a 10000 fold speed increse for that !It's like asking loads of people to work for you for free then only paying one of them.
NASA has become very immoral. |
|
Back to top |
|
|
mecej4
Joined: 31 Oct 2006 Posts: 1886
|
Posted: Mon May 15, 2017 12:44 pm Post subject: |
|
|
That has been the reaction of some people that I have talked to about this. The amount offered is barely enough to support one (1.00) fresh engineer/programmer for one year. What could that programmer hope to accomplish in that time with a large software package that requires a high degree of knowledge in many engineering, physics and mathematical topics? |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2816 Location: South Pole, Antarctica
|
Posted: Mon May 15, 2017 2:06 pm Post subject: |
|
|
As all here know very well I'm bad at English, and I might be wrong deciphering the article to my language but I did not like the sentence "The software was developed in the 1980s and is written in an older computer programming language called Fortran". The code is written in modern Fortran, and that also means the code is probably fastest within its method and hardware. But the article leaves the impression that it is slow because of some ancient language was used. Typical bs for dumb journos writing about science. It's like me writing about ballet. Respectable journals now have users' comments sections. If they don't that means the journal is an ad and propaganda place, when I go to such journals I know they will try to wash me in the toilet. That takes place everywhere, in the North and South poles, but very often after swimming here and there you will find the truth. Often somewhere near the Equator
If the entire such large code will be rewritten to other language besides Fortran most probably it will be lost becoming unmanageable just because will inflate it's already large size manyfold. Some critical utilities of course can be rewritten to other languages, C, even assembler, it's ok, but the whole large code will be dead unless it will be written and managed just with one person or few people who work as a team and most important that no single person of this team has to leave for the end of life of this code (unlikely)
As to the prize amount, they require 10x minimum speedup. For 1000x I agree I'd demand $1 M at least because realistically they most probably will need to spend $10-100M at least on new hardware
The factor of 2 in computing costs nothing. I have heard of two cases when codes (molecular dynamics and monte carlo codes) were speedup 10000 times. Calling these people to ask how, left me so far no return responds. As to speeding up 10x all here have seen that in this usergroups a month ago. I offered even less, $100, and you know what? Mecej4 politely did not take them. If I were NASA manager or a billionaire (let's hope our children will be) I'd definitely pay him much more so that he would less likely to refuse. |
|
Back to top |
|
|
JohnCampbell
Joined: 16 Feb 2006 Posts: 2554 Location: Sydney
|
Posted: Tue May 16, 2017 1:46 am Post subject: |
|
|
What I don't understand is how can you achieve these speed improvements without knowledge and access to the actual hardware configuration.
I would expect for a real solution that the ACC/GPU path or of massively distributed co-arrays would be required; both of which most would not have access to.
It must be more a PR stunt than an actual project ? |
|
Back to top |
|
|
DanRRight
Joined: 10 Mar 2008 Posts: 2816 Location: South Pole, Antarctica
|
Posted: Sun May 21, 2017 7:53 pm Post subject: |
|
|
Noticed how world changed? 20 years ago all 5 in the Top 500 Supercomputers list were in the Lawrence Livermore Lab. Look today. Top two are not only are not LLNL, not only not US they are not even the western by huge performance margin, all are Chinese.
Same pattern was with the 2016 Gordon Bell supercomputing prize. Livermore guys improved the execution speed from two years to 90 seconds and even did not get any award. Which by the way was ... $10K . Among 6 works initially selected before the final completion three were Chinese, one US, one UK and one Japanese team. Won Chinese team. |
|
Back to top |
|
|
John-Silver
Joined: 30 Jul 2013 Posts: 1520 Location: Aerospace Valley
|
Posted: Thu May 25, 2017 1:14 am Post subject: |
|
|
I inferred from the description that it's F77 code which needs updating to F95/2003/2008.
CFD by definition computer intensive, borne out by using a Plaeides SC to run it on !
I guess it's a case of working out speed improvements on a bog-standard PC or workstation and then hoping it scales up.
I wonder who the judge is.
Note also it's a pitance divided by 2 !!!
o Itwill probably b enough t buy a hotdog & a hamburger each by the time it reaches it's goal ! |
|
Back to top |
|
|
mecej4
Joined: 31 Oct 2006 Posts: 1886
|
Posted: Thu May 25, 2017 1:46 am Post subject: Re: |
|
|
John-Silver wrote: |
Note also it's a pitance divided by 2 !!!
o Itwill probably b enough t buy a hotdog & a hamburger each by the time it reaches it's goal ! |
Here is another measure:
The US Pentagon spends on average $220,000 per dead ISIS fighter, by one account (http://www.zerohedge.com/news/2017-02-09/cost-war-against-isis-reaches-11-billion). |
|
Back to top |
|
|
LitusSaxonicum
Joined: 23 Aug 2005 Posts: 2388 Location: Yateley, Hants, UK
|
Posted: Thu May 25, 2017 9:54 am Post subject: |
|
|
I'm sure there are folks who would rate that as money well spent.
You could adduce similar figures for most wars in history. What price on stopping people being herded into death camps? I'm sure that the group concession rail fares were cheaper than what it took to stop them.
More to the point, it's quite likely that rewriting the codes in a more modern language would slow them down rather than speed them up. |
|
Back to top |
|
|
John-Silver
Joined: 30 Jul 2013 Posts: 1520 Location: Aerospace Valley
|
Posted: Thu May 25, 2017 7:36 pm Post subject: |
|
|
... not to mention that people would just build bigger models just becasue 'they can' run them !!!! |
|
Back to top |
|
|
|